Introduction to epidemiological
study design
Study = basic tool in epidemiology
“An epidemiological study is a statistical study
on human populations, which attempts to link
human health effects to a specified cause”
(wikipedia.org).
• Epidemiology studies populations, not individuals
• Statistical study: requires large number of people
• Effects: often means associations but here it means
consequences
(i.e. disease, health condition)
• Cause: often means risk factor, because cause implies
causal association which is very difficult to demonstrate in
epidemiology
Epidemiology = comparison
 550 cases of stomach cancer
Epidemiology = comparison
 550 cases of stomach cancer in
Hertfordshire in 2005
Epidemiology = comparison
 550 cases of stomach cancer in
Hertfordshire in 2005
 Population 550,000
 Rate 100/100,000
Stomach cancer by age group, 2005, per
100,000
0
50
100
150
200
250
300
350
400
450
500
<25 25-34 35-44 45-54 55-64 65-74 75+
Stomach cancer in Hertfordshire,
1950-2005, per 100,000
0
50
100
150
200
250
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Stomach cancer in SE England in 2005, per
100,000
0
20
40
60
80
100
120
140
160
180
G
reaterLondonInnerLondon
Essex
SussexHertfordshire
M
iddlesex
Kent
Another example
Adult prevalence by BMI status
Health Survey for England (2008-2010 average)
Adult (aged 16+) BMI thresholds
Underweight: <18.5kg/m2
Healthy weight: 18.5 to <25kg/m2
Overweight: 25 to <30kg/m2
Obese: ≥30kg/m2
© NOO 2012
Healthy weight
40.8%
Underweight
2.1%
Overweight
32.2%
Obese
24.9%
Women
Healthy weight
31.8%
Underweight
1.7%Overweight
42.4%
Obese
24.1%
Men
Adult obesity prevalence by age and sex
Health Survey for England 2008-2010
© NOO 2012 Adult (aged 16+) obesity: BMI ≥ 30kg/m2
8.8%
16.8%
25.0%
33.1%
34.2%
30.2%
23.8%
13.9%
18.8%
25.2%
28.5%
30.6%
33.6%
26.3%
16-24 25-34 35-44 45-54 55-64 65-74 75+
Males FemalesMen Women
Adult obesity prevalence modelled estimates
National Centre for Social Research, 2006-2008
© NOO Adult (aged 16+) obesity: BMI ≥
London inset:
Adult obesity prevalence (%) 2006-2008
by local authority
13.1 to 22.3%
22.4 to 23.6%
23.7 to 25.0%
25.1 to 26.8%
26.9 to 32.9%
Obesity prevalence (%)
by Local Authority
© Crown Copyright. All rights reserved. DH 100020290 2011
Trend in raised waist circumference among adults
Health Survey for England, 1993 - 2010
© NOO 2012
The chart shows 95% confidence limits
Adults aged 16+ years
Raised waist circumference defined as >102cm for men and >88cm for women
0%
10%
20%
30%
40%
50%
Women
Men
Epidemiology = comparison
 Type of comparison (= type of study)
depends on purpose.
 E.g.
◦ Describe the disease / condition
◦ Study (analyse) its determinants / causes
◦ Study (analyse) prevention / treatment
Two primary criteria
 Descriptive vs. analytical
 Observational vs. interventional
Descriptive vs. analytical studies
 describe a pattern of occurrence of a
disease: descriptive studies (always
observational).
 to analyse the relationship between a
disease and an exposure of interest:
analytical studies (can be both
observational and interventional)
Descriptive studies
 Describe patterns of disease occurrence
 Useful for:
◦ health services planning
◦ hypothesis formulation in research
 Usually based on existing data:
◦ Mortality
◦ reporting of diseases (infections, STDs, cancers...)
◦ hospital and medical records
◦ Census
◦ employment statistics etc
Descriptive studies
4 Ws :What? Who? Where? When?
What? …… health outcome / case / event
- Mortality
- Dental health
- Chronic disease
- Cognitive function
Descriptive studies
4 Ws :What? Who? Where? When?
Person (Who?)
Age, sex, marital status, social class ….
Place (Where?)
Regions (disease atlases), internationally (Japan vs.
USA)
Time (When?)
When events occurred:
● sudden onset of diseases
● seasonal pattern (births, deaths, infections)
● secular trends
All in
relation
to the
“What”
Analytical studies
 Analysed relationship between exposure
and disease
 Often used in aetiological research
 Include
 ecological studies
 cross-sectional studies
 cohort studies
 case-control studies
 interventional studies (RCT, prevention trials etc)
Analytical studies
Analytical
studies
Observational
Ecological
Cross-
sectional
Cohort Case-control
Interventional
Randomised
control trial
Community
interventions
Population based Individual based Individual based Population based
Observational vs. interventional studies
 Observational studies are studies which observe
the populations or individuals under study; they
normally include:
 descriptive studies
 ecological studies
 cross-sectional studies
 cohort studies
 case-control studies
 Interventional studies are those where the
investigators intervene, e.g. they assign exposure
or a health measure to a particular individuals or
groups.They include:
 Prevention studies
 Randomised clinical trials
 Community interventions
Cross-sectional studies
Example
0
5
10
15
20
25
30
35%wheezinginlastyear
Parents smoke at home
Parents don’t smoke at home
Cross-sectional studies
 In a cross-sectional study, all information is
collected at one point in time
◦ Outcome
◦ Exposures
◦ Covariates
 Sometimes called “survey”
 Cross-sectional studies could be
descriptive or analytical
 Always observational
 The unit of analysis is the individual
Cross-sectional study
Time
Survey – all measurements
The only way to measure “exposures”
and “outcomes” is
- at the time of survey or
- retrospectively
Cross-sectional studies:Advantages
 Relatively quick, do not require follow up
 Provide a snapshot, e.g. prevalence of a
disease or a risk factor in population
 Allow examination of multiple diseases and
multiple exposures
 Can test or suggest hypotheses
Cross-sectional studies: Limitations
Time
Survey – all measurements
exposure
outcome
• What can
we say
about
relationship
between
outcome and
exposure?
• What can we
not say?
x
x
Cross-sectional studies: Limitations
 Since both disease and exposures are measured at the
same time, temporality is unclear
 Difficult to estimate past exposure, especially if it
occurred long time ago. Not ideal for studying
exposures that change over time (e.g. diet). (but no
problem with factors that are stable over time, e.g.
genetic markers.)
 Sensitive to reporting or recall bias if exposures are
subjectively reported.
 Sensitive to response rates and representativeness
if used to estimate prevalence of a condition in
population.
Representativeness
 Cross-sectional studies are often used to
estimate the frequency of a condition in a
population but it is usually impossible to
study the whole population
 The validity of such estimates depends
critically on the representativeness of the
studied sample
 Response rate also important
What if…
Prevalence in non-responders Total prevalence
(in full sample)
0 19%
25% 25%
50% 31%
75% 38%
100% 44%
75% response rate, and
prevalence of 25% in responders
Ecological studies
Ecological studies
 The unit of analysis is a group (e.g.
country, district, population etc)
 Data cannot be disaggregated to the level
of an individual.
 Also sometimes called correlation studies or
geographical studies
 Include comparisons over time (time-
series)
 Usually cheap and quick
Fish consumption and mortality
Ecological fallacy
 This is a logical fallacy in the interpretation of
statistical data where inferences about the
nature of individuals are deduced from
inference for the group to which those
individuals belong
 Extrapolation from groups to individuals
is conceptually inappropriate
 Situation when individual-level and group-level
(ecological) associations differ
 Individual data are necessary to estimate the
association at the level of the individual
Ecological fallacy – example
 Illiteracy rate and the proportion of the
population born outside the US:
 State-level correlation: -0.53 (the higher % of
immigrants, the lower the state’s average
illiteracy)
 Individual-level correlation: +0.12 (immigrants
were on average more illiterate than native
citizens)
 Immigrants tended to settle in states where
the native population was more literate.
Robinson, W.S. (1950). "Ecological Correlations and the Behavior of Individuals". American
Sociological Review; 15 (3): 351–357.
Ecological fallacy (1)
Blood pressure
Salt intake
Ecological fallacy (2)
Blood pressure
Salt intake
Ecological fallacy (3)
Blood pressure
Salt intake
Ecological fallacy (4)
Blood pressure
Salt intake
Example:The INTERSALT study
 Ecological analysis
◦ Increase in salt intake by 100 mmol/day was
associated with increase in SBP by 7.1 mm Hg
 Individual level analysis
◦ increase by 1.6 mm Hg of SBP
From Elliott et al, BMJ 1996
Time-series studies
 Studies repeated over time
 But not on the same individuals (i.e. not
longitudinal)
 Type of ecological studies because subjects /
events / exposures are grouped by a time interval,
hard to disaggregate individuals
 For example, health survey on a representative
sample repeated every 10 years… individual data
collected but not on the same individuals at each
survey
 They are useful for comparing changes over time
Time-series studies: use
 Compare changes over time
 Descriptive: changes in a condition over time in a
population
 Analytical: relate changes in exposure to changes
in outcome
 Long-term trends (e.g. lung cancer mortality and
smoking rates)
 Short-term variation (e.g. daily changes in air
pollution and mortality).
Time-series (vs. other ecological) studies
 Advantages
◦ help reduce confounding (e.g. it is unlikely that
smoking rates would change within a population
over a period of several days).
◦ Resemble experiment (before and after)
 Disadvantages
◦ There can be other factors changing over time -
confounding
◦ Many exposures influence health with a lag
which is often unknown (e.g. pollution and
mortality) or very long (e.g. lung cancer and
smoking).
Retention of 21+ natural teeth (%):Adult
Dental Health Surveys
Fuller E, Steele JG, Watt RG, Nuttall N. Oral health and function – a report from the Adult
Dental Health Survey 2009 (www.ic.nhs.uk)
Daily deaths and pollution
From Wichman et al, HEI research report, 2000
Ecological studies: Advantages
 Use existing (often routinely collected) data
 Quick and cheap
 Useful to general hypotheses
 Differences in both exposure and outcome rates may be
large, which increases the likelihood to find an association
 Some exposures are difficult to measure in individuals and
area-based measures are used instead (e.g. air pollution),
and some exposures are inherently ecological (e.g. income
inequality)
 Using both ecological and individual level data requires a
special type of multi-level analyses
Ecological studies: Disadvantages
 Confounding: the groups, which are compared (e.g.
countries) usually differ in many other factors than the
exposure of interest. It is often impossible to reliably
control for confounders.
 There can be systematic differences in measurements of
exposures and diseases (e.g. coding of causes of death)
between populations.
 Boundaries of different units are sometimes artificial →
misleading results.
 Ecological fallacy: ecological studies compare groups but
results are extrapolated to individuals.